Technological Strategies for Efficient Medical Data Retrieval in Interconnected Healthcare Systems: A Review

Abstract

Nowadays, the management of digital medical images faces increasing challenges due to the volume, diversity, and need for interoperability between systems. The DICOM standard has become the main format for storing and transmitting medical images, enabling the integration of data from studies such as MRI, CT, and USG. However, its complex structure and the increasing volume of data generated by interconnected devices, including IoT sensors, demand new strategies for efficient storage and retrieval. Clinical databases must support large volumes of heterogeneous data while ensuring fast access, availability, and secure information exchange. This review explores the integration of the DICOM standard with medical database systems, emphasizing the role of sensors as a primary source in clinical data management. The findings aim to support the development of more effective strategies for data retrieval and exchange, such as database fragmentation, to reduce query response times and improve information systems used by healthcare professionals and patients. Additionally, various data sets or benchmarks used in the analyzed studies are described. As a result, two approaches are identified as particularly noteworthy among the reviewed works, serving as a reference for future applications and technological developments in healthcare.

1. Introduction

The digital growth of medicine has placed sensors at the heart of the generation, integration, and exploitation of large volumes of clinical data. In the field of IoT, millions of portable and wearable devices continuously capture physiological variables [1]. Health sensors are smart devices, either wearable or implantable, used for health monitoring. Current advances in wireless communications, coupled with the rapid development of smart biosensors, have enabled continuous and cost-effective monitoring of patient health status for home rehabilitation purposes, as well as the secure transfer of individual health information via wireless networks, directly to a central data repository for clinical evaluation and subsequent analysis [2]. The recovered data enrich electronic health records, promote early detection of adverse events, and lay the foundation for personalized medicine [3]. However, sensors transcend the IoT ecosystem and are fully integrated into high-performance diagnostic equipment that produces highly complex images and signals [4].

Medical data types range from biosignals and medical images to laboratory tests and omics data [2]. Modalities such as digital radiography, CT, MRI, or even flow cytometry leverage increasingly precise sensors, generating information in DICOM, XML, or XSD format with metadata rich in clinical context [5]. The challenge is no longer just capture, but also semantic interoperability and data engineering that allow these heterogeneous sources to be merged into clinical repositories.

This work undertakes an in-depth review of recent literature to highlight the pivotal role sensors play in medical databases, including both IoT environments and traditional diagnostic devices. Particular emphasis is placed on methods that implement strategies to optimize the performance of such databases while adhering to the DICOM standard. Furthermore, the article details the distinctive attributes of each study reviewed, including: (i) whether all components required for full implementation are provided, (ii) the ease with which the proposed features can be deployed, (iii) the specific strategy adopted to enhance database performance, (iv) the technologies utilized during implementation, and (v) the benchmarks employed in the evaluation stage.

Several systematic reviews have addressed related subjects in the domain of healthcare information management. Adewole et al. [6] conducted a comprehensive meta-data analysis of CDRs, examining critical aspects such as architecture, security, AI/ML integration, and regulatory frameworks across African and global contexts. Nevertheless, this work does not address sensor-based data generation or the role of the DICOM standard in clinical database optimization. Edayan et al. [7] performed a systematic review of integration technologies in laboratory information systems, focusing on S2S communication protocols and interoperability frameworks such as HL7 and FHIR, without considering the particular challenges posed by high-volume medical imaging data management. Similarly, Tummers et al. [8] identified features and obstacles of health information systems across multiple stakeholder domains, yet their analysis remains at a high level of abstraction, without addressing specific storage or retrieval strategies for DICOM-formatted data. In contrast, the present work uniquely focuses on the convergence of the DICOM standard, sensor-generated clinical data, and database optimization strategies, particularly fragmentation techniques, offering a more targeted and technically grounded analysis of the current state of medical data management in interconnected healthcare environments.

This article is relevant because it provides an in-depth analysis of cutting-edge work in key areas of medical data storage. Beyond outlining current challenges and advances in this field, it offers a focused assessment of opportunities for developing new methods to improve database performance.

This paper is structured as follows. Section 2 introduces the materials and methods used to conduct the review. Section 3 includes the results regarding the research questions. Section 4 describes the findings, expanding the discussion on the analysis of the works. Finally, Section 5 includes the conclusions of this research and future directions.

2. Materials and Methods

This research is based on the methodological guidelines and recommendations proposed by Page [9] through the PRISMA declaration for the preparation and transparent reporting of systematic reviews, on the conceptual framework of Arksey and O’Malley [10] that defines the essential stages for conducting scoping reviews, and on the operational improvements suggested by Levac, Colquhoun, and O’Brien [11], who refine this framework by detailing iterative processes of searching, extraction, and consultation with interested parties. The authors in [10] present five well-defined steps: (1) defining the research questions, (2) identifying relevant studies, (3) applying explicit inclusion-exclusion criteria to select studies, (4) extracting and charting the data, and (5) collating, summarizing, and reporting the results.

2.1. Research Questions

Seven research questions are presented that support this work, defining the way in which the approaches included in this state-of-the-art review are evaluated, and enrich the discussion regarding the results obtained.

• RQ1. Which databases are most commonly employed to store medical information?
• RQ2. What technologies are used to view or store images in DICOM format?
• RQ3. What strategies are used to improve the performance of medical databases?
• RQ4. What data sets are most commonly used to evaluate approaches dedicated to medical data management?
• RQ5. How is medical information transmitted from sensors included in wearables and imaging machines to a central data repository?
• RQ6. What strategies are used for storing and viewing images in DICOM format in heterogeneous repositories, such as electronic medical records?
• RQ7. What are the main challenges and areas of opportunity in the current medical database landscape for future developments?

2.2. Inclusion and Exclusion Criteria

For the preliminary identification of studies, the specialized repositories that would comprise the search universe were initially defined: IEEE Xplore Digital Library, ScienceDirect (Elsevier), SpringerLink, ACM Digital Library, and MDPI. To expand coverage and detect additional potentially relevant literature, the strategy was complemented with a search in Google Scholar. All searches were conducted using the selected keywords and restricted to the time interval between 2015 and 2025, thus ensuring the timeliness and relevance of the works considered. Although the search was restricted to 2015–2025, to ensure currency, it is acknowledged that foundational database architectures developed before this period, such as early PACS implementations and relational DBMS designs, remain widely deployed in clinical environments. Where relevant, such technologies are discussed in the context of the studies that reference or build upon them, ensuring that historically established approaches are not overlooked in the analysis. The main queries used in all repositories were:

• ‘DICOM’ AND ‘database’ AND ‘sensor’ AND ‘fragmentation’
• ‘DICOM’ AND ‘database’ AND ‘sensor’

It is noted that the keyword ‘fragmentation’ was included in the first query, since, although all the techniques used in the included research will be addressed, the impact of this technique on improving database performance in the analyzed approaches will be observed.

Table 1. Keywords and related concepts.

Area	Keywords	Related Concepts
Medical databases	DICOM database sensor fragmentation	DICOM Standard Medical wearable sensors Sensors in medical imaging Biosignal sensors Data interoperability Database fragmentation Database partitioning Database sharding Horizontal, vertical, and hybrid fragmentation

presents the keywords used and the related concepts that guided the search, not only so that these words were included textually, but also mentioned in the desired context.

2.3. Relevant Studies and Application of Selection Criteria

The initial search in the specialized databases retrieved 526 records: 63 from SpringerLink, 46 from ScienceDirect, 69 from the ACM Digital Library, 13 from MDPI, and 335 from IEEE Xplore. To broaden coverage, a complementary search was conducted in Google Scholar, which generated 1090 pages of results (about 10,900 records). For feasibility, only the first 10 pages (102 articles), ordered by publication date, were screened. This screening revealed that all records previously identified in the specialized databases also appeared in Google Scholar, confirming the thoroughness of the initial search. After excluding 14 duplicates, the 102 additional articles from Google Scholar expanded the corpus to 614 unique studies.

The 614 studies were carefully reviewed to discard articles that did not include the related concepts, obtaining 160 studies. A new filter was performed, eliminating studies that did cover the concepts of interest, but whose approaches were related to other topics outside the scope of this research. Only 85 were retained and are included in the scope of this research. Figure 1 describes the stages that filtered the works found in each of the specialized repositories.

Figure 2 summarizes the annual evolution of the analyzed corpus (2015–2025) using a stacked histogram that breaks down publications by study type. The decision to include journal articles, conference papers, and book chapters reflects the publication landscape of this field: conference proceedings often represent the first formal presentation of innovative technical approaches, while book chapters provide consolidated overviews that complement primary research. Restricting the corpus to journal articles alone would have excluded relevant contributions that, in some cases, preceded or informed subsequent publications. Regarding methodological quality, a general difference was observed: journal articles tend to provide more comprehensive experimental evaluations, while conference papers and book chapters more frequently report prototypes or proof-of-concept implementations with limited validation. This distinction is reflected in the benchmark column of

Table 2. Comparison of the aspects of the main studies obtained from IEEE Xplore.

Study Reference	Technologies	Strategies to Improve the Performance	Benchmark	Data Transmission Technique	Future Research	Strategies for Heterogeneous Data
Saweros & Song [12]	Tangle (IOTA), PHR, EHR systems.	Uses DAG-based ledger to reduce latency and improve data integrity.	Evaluated through prototype implementation.	Focuses on document exchange; does not handle imaging sensors.	Smart contracts, IoT integration, and scalability across institutions.	System links structured health documents via secure channels.
Pole & Shriram [13]	3D CT images, YUV conversion, segmentation.	Compression optimized.	Tested with real CT image sequences.	Uses CT data sets; no direct sensor connectivity is described.	Improve the balance between compression ratio and diagnostic quality.	DICOM CT scans converted to YUV and compressed; metadata managed separately.
Sreepathy et al. [14]	Apache NiFi, Kafka, HDFS, Avro, JSON.	Improves scalability and flexibility with pipeline orchestration.	Evaluated with synthetic data and various ingestion scenarios.	Supports ingestion from diverse sources.	Domain-specific adapters and real-time ingestion capabilities.	The focus is on metadata and processing.
Ismail et al. [15]	Blockchain, smart contracts, PBFT, healthcare records.	Outperforms client-server in integrity and traceability.	Simulated health records.	Focuses on administrative and clinical records.	Real-time integration and broader data type compatibility.	Records managed via blockchain with smart contract control.
Latif et al. [16]	Mobile apps, SMS, cloud storage, Android platforms.	Simplifies communication through low-cost mobile solutions.	No performance benchmarks included.	Handles basic health data via mobile input.	Infrastructure expansion and interoperability standards.	Data is handled through text-based mobile health systems.
Pedrosa et al. [17]	Distributed DBs, blockchain, EHR.	Combines private blockchain and consensus layers.	Prototype tested on simulated patient records.	Handles structured EHR data.	Real-world deployment and higher fault tolerance.	Focus is on fast, secure sharing of text-based records.
Almeida et al. [18]	NoSQL, distributed storage.	Improves scalability and fault tolerance via NoSQL partitioning and replication.	Benchmarks include load time and query speed.	DICOM files ingested from PACS.	Indexing optimization and hybrid cloud deployment.	DICOM stored in Cassandra nodes; access managed via distributed queries.
Sondur et al. [19]	Edge devices, cloud storage, cache layers, IoT gateways, metadata synchronization.	Latency is reduced using local write-back caches.	Tested on synthetic workloads.	Applies to general CPS and IoT systems.	Medical-grade implementations and data reliability testing.	Architecture supports generic file and stream data storage.
Dhayne et al. [20]	EHR, HL7, FHIR, data lakes, ontologies, semantic integration.	Reviews integration architectures and semantic frameworks for big health data.	Survey-based; no experimental data sets or benchmarks analyzed.	Standard MRI and X-Ray communication.	Unified standards, data quality, and smart analytics.	Integration centered on textual and structured medical data.
Conte et al. [21]	ICT platform, data harmonization, semantic services.	The system integrates diverse inputs into standardized records.	Validated through pilot implementations across health and care institutions.	Focus on structured clinical data.	National-level integration and patient-centered services.	System processes structured records with semantic mapping.
Ourahmoune et al. [22]	Kinect, ultrasound, pose estimation, 3D tracking, knee imaging.	Combines Kinect depth data with ultrasound frames.	Tested with healthy volunteers; data set built for teaching and analysis.	Uses ultrasound sensors with Kinect for spatial reference.	Improving tracking accuracy and data set expansion.	Ultrasound data linked with Kinect pose data; stored for educational use.
Nguyen et al. [23]	Telemedicine, web viewer, smart TV, mobile interface.	Provides remote DICOM access and visualization across multiple platforms.	Prototype validated on test cases; qualitative feedback reported.	Uses existing DICOM studies.	Performance optimization and user customization.	DICOM files are viewed via web-based tools on smart TVs and mobile devices.
Ullah et al. [24]	IoT, wearable sensors, cloud storage, patient monitoring, alert systems.	Enables real-time monitoring and alerts through connected health devices.	Descriptive study based on literature and practical implementation cases.	Wearable sensors.	Expanding remote care and intelligent health alerts.	Emphasis on physiological monitoring via IoT.
Spinsante et al. [25]	IoT, AAL, sensor networks, cloud storage.	Implements modular data architecture for heterogeneous sensor data collection.	Case study with real AAL deployment; no formal benchmarks used.	Handles ambient sensors (motion, temperature).	Integration with clinical systems and AI support.	System processes contextual sensor streams for elder care.
Godinho et al. [26]	HL7, ECG, imaging modalities.	Streamlines data flow by linking imaging and clinical systems into unified reports.	Implemented in a cardiology lab (case-based).	Uses DICOM images, ECG signals, and clinical data.	Enhancing automation and expanding data standardization.	DICOM and clinical data are integrated in the warehouse.

,

Table 3. Comparison of the aspects of the main studies obtained from SpringerLink.

Study Reference	Technologies	Strategies to Improve the Performance	Benchmark	Data Transmission Technique	Future Research	Strategies for Heterogeneous Data
Moreira et al. [1]	IoT, SAREF, ontologies, semantic web, FHIR, wearable sensors.	An ontology-based model maps sensor data.	Validated via use cases and semantic reasoning tests.	Supports wearable health sensors.	Expanding domain coverage and integration with EHRs.	IoT measurements are semantically mapped using healthcare ontologies.
Syed et al. [27]	HDFS, Apache Mahout, Hadoop MapReduce.	Use of Big Data Analytics.	Physical Activity Monitoring Dataset.	WiFi to cloud servers, using an IoT-based infrastructure.	Apply to other health monitoring contexts.	Data integration in cloud environments and HDFS.
Samson & Swetha [28]	AAL, MATLAB R2018b, Raspberry Pi.	Include servers capable of signal processing.	CARDIODAT of PTB and PhysioNet.	Raspberry Pi that controls and transmits data via WiFi.	Improve the algorithm and adapt it for distributed environments.	DICOM data is stored alongside electronic medical records.
Praveenkumar et al. [29]	CR network, LSIC, DGT.	Applies hashing to detect tampering during wireless transfer.	Simulated DICOM images and wireless transmission scenarios.	Focus on DICOM image transfer; does not include direct sensor data acquisition.	Future directions involve real-time deployment and energy efficiency.	DICOM images are watermarked and transmitted securely.
Ndlovu et al. [30]	OpenMRS, DHIS2, HL7, FHIR, SMS gateways.	Proposes middleware to bridge eRecord systems and mobile apps with HL7.	Based on stakeholder interviews and implementation insights.	Focuses on text-based mobile health data; no imaging modalities or sensor integration.	National health data exchange and broader adoption of mHealth standards.	Work centered on structured alphanumeric health records.
Estrela et al. [31]	CBCT, 3D imaging, segmentation algorithms.	Enhances detection through layered filtering.	Evaluated with anonymized CBCT data sets.	Data acquisition comes from dental imaging devices.	Real-time analysis integration.	Metadata and segmentation are used for detection.
Napravnik et al. [32]	NLP, multimodal database, unsupervised learning, CNN, PACS.	Uses image-text embedding and weak supervision to auto-label large data sets.	Developed using 18M images from PACS and 12M associated radiology reports.	Includes CT, MRI, and X-Ray images.	Refining embeddings and cross-center generalization.	DICOM images linked to reports via NLP; annotations built through embeddings.
Safaei & HabibiAsl [33]	Vertical fragmentation, indexing engine, query optimizer.	Reduces search time using fragmented indexing and parallel query processing.	Validated with hospital PACS image data sets; retrieval time benchmarks reported.	DICOM images from hospital systems; analysis occurs post-acquisition.	Cloud integration and dynamic indexing strategies.	DICOM metadata indexed by attributes; retrieval guided by feature similarity.
Safaei [34]	Hybrid fragmentation, metadata indexing, feature clustering, Apache Lucene.	Combines vertical and horizontal fragmentation.	Tested on PACS data sets; retrieval speed and accuracy compared to baseline.	Uses DICOM images; assumes acquisition from hospital imaging systems.	Scaling fragmentation to big data and cloud platforms.	DICOM attributes split by access patterns; clustered features aid retrieval.
Pagella Aguero [35]	AI models, CT/MRI/X-Ray.	Enhanced preprocessing and segmentation.	Does not include evaluation on specific data sets.	Covers CT, MRI, and X-Ray images.	AI integration, real-time analysis, and smart imaging.	DICOM is used for clinical input.
Elloumi et al. [36]	CT, 3D segmentation, thresholding, tumor volume estimation.	Uses voxel analysis and 3D surface generation to identify tumor regions.	Tested on CT scan data sets; evaluated using segmentation accuracy.	CT images used as input; data processed post-acquisition.	Future work includes refinement of 3D modeling and clinical integration.	DICOM CT data segmented into 3D volumes.
Blazek et al. [37]	Laboratory IS, workflow engine, encryption, access control, backup systems.	Security is reinforced through encryption layers and workflow control policies.	No standard benchmarks used.	Includes a module for connecting to external data sources.	AI integration and broader interoperability across the hospital system.	The system manages structured lab records with secure access.

,

Table 4. Comparison of the aspects of the main studies obtained from ScienceDirect.

Study Reference	Technologies	Strategies to Improve the Performance	Benchmark	Data Transmission Technique	Future Research	Strategies for Heterogeneous Data
Pezoulas et al. [2]	IoT devices, wearable sensors, EHR systems, mobile apps.	Performance depend on integration of heterogeneous data sources.	No test or benchmark data set is mentioned.	Focuses on data classification, not transmission workflows.	Standardization of formats.	Data integration across structured and unstructured repositories.
Rinty et al. [38]	GNU Health, Tryton, CDA.	Distributed databases by sites, loose coupling between servers.	A test data set with 100 patients, 20 healthcare professionals, and 5 care centers was used.	Transmission occurs via Tryton, which acts as an interface.	Expanding to a real-time system with IoT and machine learning integration.	DICOM images are integrated into the CDA documents.
Akhtar et al. [39]	AI, ICT, CPS, IoT, and blockchain.	Optimization of compression algorithms.	No test or benchmark data set is mentioned.	QoS-aware middleware, Fiber optic sensor.	Creating scalable and affordable models for developing countries.	DICOM-optimized compression techniques.
González et al. [40]	Distributed data lake, mobile health services, IoT.	Microservices organized by layers and use of distributed databases.	No test or benchmark data set is mentioned.	Sensors connected to smartphones.	A methodology for deployment in real-life clinical settings.	It stores both types of information separately.
Wang and Nurcahyo [41]	BigchainDB, IPFS, Blockchain.	Using BigchainDB instead of traditional databases.	No test or benchmark data set is mentioned.	IoT devices and wearables transmitting data to mobile apps.	Integration with proprietary devices and hospital EHRs.	Medical images and alphanumeric information are separated.
Vesselkov et al. [42]	Cloud infrastructure, EHR systems, data analytics tools.	Performance enhanced via service modularity.	No specific data sets used.	Focus on teleconsultation systems.	Data governance and platform interoperability.	No DICOM-specific storage discussed.
Loomis et al. [43]	Dental radiographs, panoramic X-Ray, postmortem imaging, forensic software.	Performance linked to radiograph quality and anatomical variations.	No benchmarks used.	No real-time sensor transmission involved.	Suggests the need for standardized acquisition protocols.	DICOM used with annotated forensic dental data.
Beier et al. [44]	EDF, XML, OpenStack Cloud, MATLAB, secure FTP.	Uses federated access control and standardized formats.	Includes real data sets from multiple labs across Europe.	Sleep data from EEG, ECG, oximetry, etc.	Scaling to more centers, adding AI tools, and enhancing data harmonization.	Physiological signals stored in EDF and metadata in PostgreSQL.
Yang et al. [45]	HDFS, MapReduce, cloud infrastructure.	Improves image access time.	Simulated DICOM workloads in cloud HDFS clusters.	It is assumed that DICOM files are obtained from various sources	Hybrid cloud integration and dynamic replica adjustment.	DICOM images stored in HDFS.
Chahal et al. [46]	MRI, CNN, fuzzy logic, segmentation.	Combines CNN and fuzzy logic for improved tumor detection accuracy.	Tested on MRI data sets.	MRI images used as input; processing applied post-acquisition.	Real-time deployment and clinical validation.	Annotations stored separately.
Enzmann et al. [47]	PACS, app-based architecture, modular platforms.	Advocates modular, API-driven systems for flexible and scalable imaging workflows.	Descriptive proposal based on comparison with smartphone ecosystems.	Applies to radiology imaging.	Future vision includes app ecosystems, plug-and-play tools, and unified UIs.	DICOM is central to the proposed framework and is accessed via modular app-based layers.
Zheng [48]	Digital radiography, enhancement filters, image fusion.	Highlights image-clarity improvements through preprocessing and hybrid techniques.	Review of current applications; no benchmark data sets evaluated.	Covers X-Ray, CT, and MRI images; focuses on post-acquisition processing.	Automation, AI-assisted diagnostics and fusion techniques.	DICOM enhanced with preprocessing and segmentation tools.

,

Table 5. Comparison of the aspects of the main studies obtained from ACM Digital Library.

Study Reference	Technologies	Strategies to Improve the Performance	Benchmark	Data Transmission Technique	Future Research	Strategies for Heterogeneous Data
Vázquez-Ingelmo et al. [49]	Django, OpenCV, TensorFlow, 3D Slicer, Cornerstone.js.	Optimized image retrieval and metadata indexing improve platform performance.	Tested on anonymized CT/MRI data sets from collaborating medical institutions.	Supports the upload and processing of MRI and CT scans.	Plans include integration with clinical systems.	DICOM images and patient data are stored in PostgreSQL.
Song et al. [50]	CDA, XML, Dropbox.	Uses standardized document formats.	Evaluation conducted with test cases using a sample of PHR and imaging documents.	Supports upload of DICOM files.	Mobile integration and patient engagement tools.	DICOM data linked to health records via HL7 CDA.
Galletta et al. [51]	Cloud storage, XML, multi-cloud storage, HIS.	Improved performance by allocating fragments across different servers	Tested with real MRI data sets from a hospital’s radiology department.	MRI data collected and distributed across cloud sites; assumes PACS for acquisition.	Improved query latency and larger-scale deployment.	DICOM MRI files stored in Cloud providers.
Singh et al. [52]	Blockchain, access control, encryption, cloud storage.	Highlights secure architecture models with layered access and integrity checks.	Survey-based; no experimental benchmarks or data sets evaluated.	Discusses general data sources.	Dynamic access models, encryption improvements, and policy frameworks.	Focuses on general secure storage and access practices.
Gleiss & Lewandowski [53]	EHR, workflow tools, digital platforms, HDMP, interoperability.	Highlights organizational duality to achieve balance between innovation and routine IT use.	Based on case studies in hospitals, no benchmarks or data sets were used.	Focuses on administrative systems.	Adaptive IT strategies and policy co-design.	Emphasis on digital governance and IT culture in health.

,

Table 6. Comparison of the aspects of the main studies obtained from MDPI.

Study Reference	Technologies	Strategies to Improve the Performance	Benchmark	Data Transmission Technique	Future Research	Strategies for Heterogeneous Data
Ait Abdelouahid et al. [54]	HL7 FHIR, CDA, openEHR.	Interoperability improved through standard mapping and semantic models.	No benchmark data sets.	Focuses on data structures and exchange standards.	Harmonization of standards, governance models, and tooling support.	DICOM was considered an external format.
Cernian et al. [55]	Blockchain (Ethereum), PHR.	Integrates PHR with wearable sensors using access controls with tokens.	Bucharest clinic database.	Does not include direct acquisition from imaging sensors.	Full EHR integration, real-time monitoring, and access transparency.	Access and integrity are managed through Ethereum.
Nowakowski & Kaczmarek [56]	Thermal cameras, AI models, CNNs, IR sensors.	AI improves classification accuracy.	Performance evaluated on thermal image data sets.	Thermal sensors are used to collect physiological data.	Multi-modal fusion, standardization, and broader clinical trials.	Thermal data is analyzed via AI pipelines for diagnostic purposes.
Koutras et al. [57]	IoMT devices, encryption, blockchain.	Layered architectures combining cryptography.	No experimental benchmarks or data sets used.	Discusses sensors in medical wearables.	Lightweight encryption.	Focus on network-level security in IoMT ecosystems.
Huda et al. [58]	Federated identity, digital healthcare, risk scoring.	Evaluates threats via probabilistic models and risk scores on identity systems.	Risk assessment applied to simulated healthcare network configurations.	Focus on authentication and access.	Broader clinical validation and dynamic risk adaptation.	The framework centers on identity and access security.
Althenayan et al. [59]	Chest X-Ray, CT, deep learning, multimodal fusion, CNN.	Combines CT and X-Ray features using a hierarchical CNN-based classifier.	Trained and tested on publicly available COVID-19 DICOM data sets.	Uses CT and X-Ray images in DICOM format.	Model generalization and clinical deployment.	DICOM images from two modalities were fused in a CNN pipeline for COVID detection.
Shakor & Khaleel [60]	Deep learning, cloud platforms, CT/MRI, federated learning.	Explores scalable AI pipelines for image analysis using cloud infrastructure.	No benchmark data sets.	Covers CT/MRI image processing; assumes acquisition from clinical systems.	Privacy-preserving AI and cross-institution learning.	DICOM images processed in cloud AI systems.
Mohsan et al. [61]	Blockchain, IPFS, patient-centric storage.	Uses smart contracts and IPFS hashes to secure and retrieve medical images.	Prototype tested on DICOM images and PDF reports.	Handles DICOM images and clinical reports; assumes external acquisition.	Fine-grained access policies and large-scale deployment.	DICOM files stored in IPFS; blockchain ensures secure access and traceability.
Khan et al. [62]	IoHT, edge devices, MATLAB, RFID, WSN.	Combines detection accuracy with secure communication protocols.	Tested with simulated sensor data and performance metrics (latency, accuracy).	Focuses on wearable and ambient sensors.	Real-time deployment and integration with medical clouds.	Targets lightweight, secure IoHT data transmission.

and

Table 7. Comparison of the aspects of the main studies obtained from Google Scholar.

Study Reference	Technologies	Strategies to Improve the Performance	Benchmark	Data Transmission Technique	Future Research	Strategies for Heterogeneous Data
Leif & Leif [5]	Flow cytometry, XML, gating metadata, MIFlowCyt.	Uses XML schemas to standardize and validate cytometry experiment metadata.	Demonstrated via metadata annotation in cytometry files; no benchmarks used.	Focuses on flow cytometry data.	Broader adoption of MIFlowCyt and schema refinement.	Metadata standardized via XML for cytometry workflows.
Le et al. [63]	Cloud computing, DTLZ, NSGA.	Vertical partitioning, Grid Partitioning.	Real DICOM data set.	Not addressed, but PACS or RIS is assumed by the use of DICOM.	Extend the application of NSGA-G to other hybrid benchmarks.	Fields extracted from DICOM files are stored in either row or column formats.
Lee et al. [64]	Blockchain, IPFS, Hyperledger Fabric.	Load reduction through modular REST services.	Synthetic data used for evaluation.	Medical files are uploaded via a web interface.	Challenges include system scalability and interoperability.	Medical images stored in IPFS.
Czelusniak et al. [65]	Agent-based systems, JADE framework.	Agents handle distributed tasks like retrieval and routing.	Prototype tested with sample DICOM images and a simulated hospital network.	Sensors like MRI/CT are assumed as sources.	Semantic enrichment, scalability, and integration with HIS systems.	DICOM images are routed via agents and annotated semantically.
Udhay et al. [66]	CT, MRI, 3D navigation system, optical tracking, planning software.	Accuracy enhanced by preoperative planning, real-time tracking, and imaging fusion.	Case-based evaluation with surgical CT and MRI data sets; no standard benchmarks.	Imaging from CT/MRI is used for navigation.	Improving precision, workflow integration, and wider clinical use.	DICOM images aligned with navigation coordinates.
Gyrard et al. [67]	FHIR, OMOP CDM, federated learning.	Uses federated queries and common data models to enable cross-border analytics.	BigPicture.	Focus on structured clinical and genomic data.	Standard alignment, legal interoperability, and AI integration.	Work focuses on structured cancer-related clinical data sets.
Schreiweis et al. [68]	FHIR, openEHR, HL7, data lake, IHE, linked data.	Combines semantic models with flexible data integration for advanced querying.	Evaluated through case studies and platform deployments.	Focuses on clinical records and metadata.	Expanding ontologies and supporting multi-center collaboration.	Structured data is handled via semantic web technologies.
Vega et al. [69]	SPARQL, biomedical ontologies, RDF, VTK.	Ontology-based annotation enhances semantic interoperability.	Prototype tested with sample DICOM images annotated.	Assumes external imaging sources.	Multi-user support and integration with clinical workflows.	Metadata structured using ontologies.
Tura et al. [70]	EHR systems, cloud platforms, PACS, HL7.	Highlights interoperability frameworks.	Analysis based on Russian healthcare.	Mentions general diagnostic devices.	National health cloud platforms and legislative alignment.	No technical integration described.
Pedrosa et al. [71]	Federated ledger, Rust, pseudonymization.	Blockchain-based identity and auditability.	Use case modeling; no data sets used.	Focuses on structured records; no use of direct sensor input.	Real-world testing and integration with EHR infrastructures.	System handles identity-protected access to clinical records.
Alyami et al. [72]	PHR systems, HL7, secure messaging, Dropbox access token.	Proposes flexible data access and communication.	No performance benchmarks used.	Focus on textual records and scheduling.	Adoption incentives, usability, and regulatory compliance.	System prioritizes patient-driven access to structured records.
Munagandla et al. [73]	Stream processing, Kafka, EHR, real-time dashboards.	Improves emergency decisions through event-driven data aggregation.	Prototype validated with simulated emergency scenarios.	Uses location and clinical status data.	Sensor data fusion and national emergency platform links.	The system integrates real-time alphanumeric health streams.
Mitrano et al. [74]	Cloud computing, WfMS, EHR, telemedicine modules, web services.	Workflow engine streamlines task coordination and medical data access.	Based on implementation insights from Italian telemedicine projects.	Focus on clinical workflows.	Improving scalability and EHR interoperability.	Platform centers on document workflows and patient data.
Gornale et al. [75]	X-Ray, image annotation, grading labels.	Structured metadata and expert labeling enhance data set usability.	Includes 8892 DICOM knee X-Rays with severity grades and metadata.	Images collected via X-Ray devices.	Clinical validation and integration with ML pipelines.	DICOM images with KL grades; structured for ML training.
Deniz & Kaya [76]	PSP sensors, PACS, artifact classification, image quality analysis.	Error types were identified via visual inspection and categorized for clinical relevance.	The data set includes 1200 intraoral PSP images.	Uses PSP-based digital radiographs.	AI-based artifact detection and imaging protocol updates.	DICOM format used; images labeled by error type for quality assurance.
Khoroshun et al. [77]	CT, RadiAnt, metal artifact reduction.	Enhances CT clarity by correcting beam hardening and metal-induced noise.	Evaluated with real CT images of gunshot cases.	CT scans are used as input.	Real-time filtering and integration with diagnostic tools.	DICOM CT data processed with correction filters.
Borovska et al. [78]	IoMT, ultrasound, Tensor Flow, AI analytics, BSN.	Combines IoMT and AI to enable remote diagnostics with low-power data links.	Evaluated with ultrasound image streams and diagnostic analytics.	Uses USG imaging via IoMT devices; real-time data sent to the cloud.	Energy efficiency, broader cancer screening, and IoMT scaling.	Ultrasound DICOM images streamed to the cloud; processed with AI diagnostic tools.
Zimmerer & Gellrich [79]	CT, MRI, CoDiagnostiX, PACS, Visage.	Combines real-time imaging and QA metrics to improve surgical outcomes.	Evaluation based on clinical implementation.	Uses CT and MRI intraoperatively.	Use of 3D fluoroscopy.	DICOM images linked to navigation and QA systems.
Satti et al. [80]	UHPr, big data, HL7, FHIR, data curation, interoperability services.	Applies rule-based pipelines and semantic mapping for unified data exchange.	Platform tested on health records from regional pilot systems.	Focuses on structured health data.	Scaling data harmonization and ontology alignment.	Platform handles curated structured data via HL7/FHIR.
Podchashynskyi et al. [81]	Compression algorithms, partial loss.	Assesses compression methods balancing file size and diagnostic fidelity.	Comparison made using video data sets.	Applies to medical video streams; DICOM format used for some evaluations.	Adaptive compression tuned to clinical requirements.	DICOM and video frames are compressed.
Molaei et al. [82]	CNN, fuzzy logic, fractal features, chest X-Ray.	Combines fuzzy CNN and fractal descriptors to enhance classification accuracy.	Tested on DICOM chest X-Ray data sets; performance measured with standard metrics.	Uses DICOM chest X-Rays.	Generalizing the model to other lung pathologies and modalities.	DICOM images are input into a hybrid AI model; features are extracted for diagnosis.
Rasheed et al. [83]	Distributed systems, EHR, blockchain, cloud storage, access control.	Outlines architectural and policy hurdles in decentralizing health systems.	No data sets used.	Focus on record-level decentralization.	Governance models, scalability, and patient access.	Study centered on distributed EHR management.
Kotzias et al. [84]	IoT, cloud computing, AI, big data, robotics, 3D printing.	Explores the integration of I4.0 technology.	No benchmark data sets.	Describes general applications, emphasizing the use of sensors.	Aligning digital innovation with clinical workflows.	Emphasis on broad digital health transformation.
Rönnau et al. [85]	openEHR, HL7, relational DBs, big data platforms.	Takes advantage of distributed engines.	Prototype validated using synthetic EHR and DICOM image data.	Handles DICOM images and HL7 messages; assumes external acquisition.	Real-time analytics and adaptive data pipelines.	DICOM data integrated with structured records via big data frameworks.
Bracciale et al. [86]	IoT, PACS, access control, cybersecurity audit.	Reveals widespread misconfigurations and weak access policies in PACS servers.	Field study of 3000+ DICOM nodes; tested for vulnerabilities and open ports.	DICOM servers scanned across hospitals.	Enforcing authentication and segmenting medical networks.	DICOM servers lacked encryption, highlighting the need for stricter controls.
Roehrs et al. [87]	PHR, EHR integration, IoT, mobile apps, patient portals.	Synthesizes models for usability, access control, and patient data sharing.	Review of 90+ studies; no benchmarks or data sets tested.	Focus on textual and structured records; imaging is rarely addressed.	Recommends design for usability, privacy, and semantic compatibility.	Review centers on structured PHR interoperability.
Tiriteu et al. [88]	ERP, HIS, digital dashboards, cloud services, data integration.	Links hospital departments via unified digital interfaces.	Case study of hospital ERP implementation; no benchmark data sets used.	Focuses on administrative and clinical data.	Predictive analytics and workflow automation.	Emphasis on hospital-wide digital infrastructure.
Ranschaert [89]	PACS, RIS, teleradiology, cloud storage, AI tools.	Describes efficiency gains from IT in image access, sharing, and reporting.	Narrative overview; no performance benchmarks or data sets evaluated.	Focuses on the post-acquisition management of radiological images.	Suggests enhancing interoperability and AI-assisted reporting systems.	DICOM is central to the workflow; integrated with PACS and reporting systems.
Huang et al. [90]	Genomics, EHR, imaging data, AI, integrative analytics.	Highlights the convergence of heterogeneous data sources.	No benchmarks or empirical data sets evaluated.	No specific data transmission technique is described.	Interoperability, data governance, and actionable insights.	DICOM referenced among modalities; integration methods not detailed.
Pendyala [91]	Cloud platform, HL7 FHIR, RDF, semantic layer, microservices, data mapping.	Applies layered architecture to enable semantic interoperability.	Validated in hospital IT context; no benchmark data sets disclosed.	Focus on structured clinical records; imaging systems are not explicitly addressed.	Integration of AI tools and wider standard adoption.	Focus is on structured data using FHIR and RDF models.
Soltanmohammadi [92]	PostgreSQL 12.3, Python 3.8, hashlib (MD5/SHA-256/SHA-3), psycopg2.	Two-layer schema: (1) MD5 hashing of the patient identifier for anonymization; (2) modulo mapping to native PostgreSQL partitions (10 test partitions)	Synthetic/dummy datasets of 1, 5, 10, and 15 million records.	Not applicable–dummy data generated directly, without sensor acquisition.	Validation with larger-scale real datasets; batch insertion, two-phase commit for updates, soft delete; dynamic partition rebalancing	Not addressed.
Dubey & Saxena [93]	MongoDB (GridFS/ BSON Binary), PyMongo, Python; DenseNet121, ResNet, MobileNet, Xception.	Document-oriented storage: binary image and metadata in a single document; metadata indexing for fast retrieval; MongoDB’s native horizontal scalability.	MIMIC-IV CXR subset: 230 normal images and 234 with pneumonia, 80/20 split; classification with DenseNet121.	No real-time transmission from sensors.	Integrate additional deep learning architectures; validate generalization across diverse conditions for real-world clinical deployment	Combines image binary and clinical metadata in a single MongoDB document.

, where the absence of formal evaluation is more prevalent among these latter source types. No language restrictions were applied during the search; however, all retrieved studies were published in English, which may reflect a publication bias rather than an intentional exclusion criterion.

At first glance of Figure 2, a gradual increase in the total volume is evident—with notable peaks in 2020 (13 studies) and 2024 (14 studies)—and a shift in the composition of sources: while in 2015–2016 conference papers predominated, from 2017 onward, journal articles clearly became the majority and concentrated the most recent production. Below the x-axis, the figure includes a numerical table that facilitates precise reading of each bar, indicating how many journal articles (green), conference papers (purple), and book chapters (orange) are recorded per year. In aggregate terms, 65.88% of the total corresponds to articles published in journals (56 of 85 studies), 29.41% to conference proceedings (25 studies), and only 4.71% to book chapters (4 studies), which confirms the relevance of the topics considered in this work and the high quality of the papers reviewed in this study.

Figure 3 summarizes the provenance of the 85 primary studies identified in the systematic search. Google Scholar accounts for the largest share of the corpus (37%), confirming its usefulness as a collector of multiple sources, replicating and expanding on the results found in specialized databases. Among the disciplinary repositories, IEEE Xplore contributes 15 works (18%), followed by SpringerLink and ScienceDirect with 12 each (14% respectively), while MDPI adds 9 records (11%), and the ACM Digital Library 5 (6%).

Figure 4 summarizes the geographical origin of the studies included in this review. As can be seen, the largest proportion of papers originated in the United States, followed by a European core led by Switzerland, the Netherlands, and the United Kingdom (between 9 and 11 studies each). The contributions of the remaining countries are much more dispersed: only Canada, India, and Ukraine contribute more than two articles, while nations in Asia, Latin America, Africa, and Eastern Europe, for example, Japan, Brazil, and Kenya are represented with one or two studies.

The predominance of studies from the United States and Europe in Figure 4 likely mirrors broader patterns in research output and database infrastructure investment, rather than a limitation of the search strategy itself. This geographical imbalance is acknowledged as a potential constraint on the global generalizability of the conclusions drawn in this review.

2.4. Collating, Summarizing, and Reporting

Once the main articles were obtained for review in this study, different characteristics of each approach were extracted for further analysis. The following sections detail the characteristics evaluated and the findings, continuing the PRISMA process.

3. Results

The studies obtained were reviewed according to six aspects derived from the research questions described above. The aspects are described below:

• Technologies used by each approach to store medical information, to view images in DICOM format, or to manage these same types of images. The technologies identified in each study are reported in the corresponding column. It should be noted, however, that PACS systems and database management systems are not included in this field, as both are addressed separately in dedicated tables within the discussion section, where a more detailed and comprehensive analysis of each is presented.
• The different strategies (if mentioned) to improve the performance of medical databases. The strategies documented in the reviewed studies that seek to optimize the performance of medical databases were analyzed.
• Benchmarks or data sets used to evaluate each approach. The review identified benchmarks or data sets used in the studies to evaluate the effectiveness and performance of the different technological strategies.
• Strategies used to transmit information from sensors to a central data repository. We reviewed the different documented communication strategies for the efficient and secure transmission of data obtained by medical sensors to central clinical information repositories.
• The challenges presented in each study or projects for future research were documented and analyzed, primarily related to technical, operational, and economic aspects.
• Strategies used for storing DICOM images in heterogeneous repositories that include alphanumeric medical information. These approaches stand out for their ability to efficiently manage diverse data in a single environment, addressing interoperability and comprehensive management of medical information in complex clinical systems.

Table 2, Table 3, Table 4, Table 5, Table 6 and Table 7 describe each of the main studies under the aspects already mentioned. The tables organize the studies by the data source from which they were obtained and collectively provide a structured overview of the 86 primary studies included in this review. Across all sources, considerable heterogeneity is observed in the technologies employed, the performance strategies adopted, and the communication mechanisms used to transmit data from sensors to central repositories. While some studies report the use of specific benchmarks or recognized data sets to validate their approaches, a notable subset relies on synthetic workloads, prototype implementations, or qualitative evaluations, reflecting the current lack of consolidated benchmarking standards in the field. Similarly, strategies for handling DICOM images within heterogeneous repositories range from distributed and cloud-based architectures to semantic metadata linkage, underscoring the diversity of solutions currently being explored. The findings extracted from these tables serve as the foundation for the in-depth analysis presented in the following section, where each research question is addressed in detail.

4. Discussion

4.1. RQ1. Which Databases Are Most Commonly Used to Store Medical Information?

The analysis of the reviewed studies reveals a notable diversity in the technologies employed for medical data storage, yet some patterns emerge. Distributed databases and Blockchain-based architectures recur across several contributions. For instance, Pedrosa et al. [17] emphasize the integration of distributed databases with blockchain to enhance data security and traceability in EHR. Similarly, Ismail et al. [15] propose the use of blockchain along with IPFS, reinforcing a trend toward decentralized and immutable storage models.

Another frequently cited component is the adoption of EHR platforms, often augmented with standards such as HL7 or ISO 13606, as seen in the study by Conte et al. [21]. These standards ensure interoperability and semantic consistency across healthcare systems. Complementary technologies like cloud storage, edge computing, and IoT devices are also integrated to support scalable and low-latency access to medical data, as demonstrated by Sondur et al. [19].

Enriching the answer to the research question,

Table 8. DBMS included in the analyzed works.

DBMS	DBMS Type	Study Reference
MongoDB	Document store	[14,18,20,21,30,40,55,93]
MySQL	Relational	[14,20,30,40,50]
CouchDB	Document store	[14,18,20,64]
PostgreSQL	Object-Relational	[14,20,38,40,92]
Cassandra	Wide-column store	[14,18,20,55]
Oracle	Relational	[14,55,63]
GraphQL	Graph/query language	[38,55,91]
SQL Server	Relational	[30,33,55]
Redis	Key-value store	[20,40]
HBase	Wide-column store	[14,20]
SAP HANA	In-memory columnar	[20,63]
MariaDB	Relational	[18]
Hypertable	Wide-column store	[20]
Neo4J	Graph (property graph)	[20]
Teradata	Relational/data warehouse	[20]
AllegroGraph	Graph (RDF/triple store)	[20]
Openlink Virtuoso	Graph (RDF/SPARQL)	[20]
Zope Object Oriented database	Object-oriented	[30]
Modex BCDB	Document/blockchain	[55]
HYRISE	In-memory columnar	[63]
AWS Neptune	Graph (managed)	[91]
AWS QLDB	Ledger/immutable	[91]

shows the use of DBMS mentioned or considered in the analyzed works.

Table 8 synthesizes the database back ends reported across the corpus and underscores the technological heterogeneity of medical data management. Conventional relational systems (MySQL, PostgreSQL, Oracle, SQL Server, MariaDB, Teradata, SAP HANA) remain foundational for transactional EHR and registry workloads. Document-oriented stores (MongoDB and CouchDB) and wide-column/BigTable-style engines (Cassandra, HBase, Hypertable) are adopted to accommodate semi-structured clinical data and high-throughput ingestion. Overall, the recurrent presence of MongoDB, MySQL, PostgreSQL, and CouchDB, alongside more specialized engines, illustrates how the analyzed works align DBMS choice with diverse data modalities (relational, document, image, and graph) and with priorities of scalability, interoperability, and performance.

Notably, the works of Safaei [33,34] stand out for their specialized focus on PACS. Safaei & HabibiAsl [33] introduce a vertical fragmentation model integrated with indexing engines and query optimizers to reduce retrieval time from PACS databases. Meanwhile, Safaei [34] proposes a hybrid fragmentation scheme combining both vertical and horizontal fragmentation, using feature clustering and metadata indexing to enhance retrieval performance. Both studies underscore the importance of tailoring database strategies to the access patterns and structural properties of medical imaging data, particularly DICOM files. Adding to this picture, Soltanmohammadi et al. [92] apply MD5-hash-based partitioning directly within PostgreSQL, demonstrating that hash-driven fragmentation of structured patient records remains a viable complement to the attribute-based fragmentation strategies discussed above, particularly for transactional RDBMS workloads rather than imaging metadata.

Overall, blockchain, EHR systems, and distributed or fragmented database models emerge as the most consistently utilized technologies. The prominence of PACS-specific approaches in Safaei’s work highlights a complementary trend in optimizing image-based medical data management.

4.2. RQ2. What Technologies Are Used to View or Store Images in DICOM Format?

A wide range of technologies is employed to manage medical images in DICOM format, encompassing both storage and visualization tools. Commonly used storage solutions include PACS, which serve as centralized repositories for DICOM images and often integrate with hospital information systems. Several works have adopted PACS or custom PACS-like systems to archive and retrieve medical images effectively (e.g., [27,32]).

For image viewing, a variety of technologies are used, including dedicated DICOM viewers and web-based tools. Some systems incorporate open-source frameworks such as Orthanc or VTK [14,69], which not only provide storage but also RESTful interfaces compliant with DICOMweb, facilitating image visualization over web applications.

The primary aim of PACS was to create a centralized digital repository for various imaging modalities such as X-ray, CT, MRI, and ultrasound, thereby streamlining radiological workflows and enhancing diagnostic efficiency [94]. Many studies use or consider PACS to manage images in DICOM format (31 studies). However, many approaches create their own PACS implementation, and others do not describe the specific PACS application they use.

Table 9. Viewers and PACS systems obtained from the analyzed works.

Technology	Study Reference	Data Storage	Visualization Capability	PACS or Viewer
Dicoogle	[18,26,71]	x	x	PACS
Own implementation	[23,45,61]	x	x	PACS
Orthanc	[14]	x	x	PACS
e-Vol DX 2.0	[31]		x	Viewer
SyngoVia	[36]	x	x	PACS
MedDApp	[41]	x		PACS
XNAT	[44]	x	x	PACS
Cornerstone.js	[49]	x	x	PACS
MicroDicom	[59]		x	Viewer
WebMedSA 3D-Visualizer	[69]		x	Viewer
Ak Dental Ltd	[76]	x		PACS
RadiAnt	[77]		x	Viewer
Medimsight	[78]	x		PACS
C-arms	[79]	x		PACS
Ziehm Vision RFD	[79]		x	Viewer
Siemens Arcadis Orbic 3D	[79]		x	Viewer
Philips Veradius Neo	[79]		x	Viewer
Brainlab Curve/Kolibri	[79]		x	Viewer
PixelMed Java DICOM Toolkit	[85]		x	Viewer

Technologies that include Data Storage and/or Visualization Capability features are marked with an “x”.

presents the studies that specifically mention the PACS they use or if they developed a new approach. It can be seen that the most widely used PACS is Dicoogle; however, most of the papers that mention the use of PACS do not describe the specific implementation included. Three articles describe their own implementation of PACS functions with different characteristics and environments.

In [23], highlighted for their contributions, the authors employ a combination of DICOMweb services and cloud-based technologies. These solutions include the use of WADO-RS for image retrieval and visualization directly through web browsers, eliminating the need for specialized desktop applications and supporting mobile access.

Additionally, cloud-based services and hybrid architectures are being explored to enhance scalability and accessibility of DICOM image handling, often integrating with IoT devices and medical sensors to streamline the data flow between acquisition and analysis platforms [60,63,69].

In summary, the technologies used to view or store images in DICOM format range from traditional PACS systems and DICOM viewers to modern, web-enabled platforms and RESTful services like DICOMweb. These solutions reflect the evolution of medical imaging infrastructure towards more flexible, interoperable, and scalable systems.

The prevalence of custom PACS implementations suggests that existing open-source solutions, while functional, do not always meet the specific technical or institutional requirements of individual research contexts, such as integration with proprietary hospital systems, support for non-standard workflows, or deployment in resource-constrained environments. This gap represents an area of opportunity for the open-source community to develop more flexible and modular PACS frameworks that can be adapted without full reimplementation.

4.3. RQ3. What Strategies Are Used to Improve the Performance of Medical Databases?

To enhance the performance of medical databases, researchers have implemented a variety of strategies focused on improving scalability, latency, throughput, and communication efficiency. One notable approach is the use of distributed ledger technologies, such as DAG-based systems, which have been shown to reduce latency and increase throughput in healthcare data exchanges (Saweros and Song [12]). Compression techniques are also employed to optimize storage and transmission, particularly in scenarios involving large imaging data sets such as CT scans (Pole and Shriram [13]).

Pipeline architectures leveraging tools such as Apache NiFi and Kafka enable better scalability and flexibility by streamlining data ingestion and processing across heterogeneous sources (Sreepathy et al. [14]). Furthermore, the adoption of decentralized storage and blockchain frameworks has demonstrated improvements in data integrity and trust when compared to traditional client-server models (Ismail et al. [15]).

Other strategies include the use of lightweight mobile applications and SMS-based systems to simplify communication, especially in low-resource settings where infrastructure may be limited (Latif et al. [16]). These approaches collectively illustrate the diversity of methods used to boost the efficiency and reliability of medical data systems, tailored to various technological and contextual constraints.

A functional comparison of the reviewed approaches reveals meaningful differences across key categories. For DICOM storage and access, Almeida et al. [18] and Galletta et al. [51] demonstrate that distributed NoSQL architectures improve scalability over traditional PACS, though at the cost of increased deployment complexity. For cross-institution data exchange, Pedrosa et al. [17] and Lee et al. [64] adopts blockchain and IPFS to enable secure sharing, while Rinty et al. [38] rely on loosely coupled distributed databases with HL7 standards; the former prioritizes auditability while the latter favors simplicity. For data integrity and access control, blockchain-based approaches such as Ismail et al. [15] and Mohsan et al. [61] outperform traditional models in traceability, but face limitations in throughput when handling large imaging files. Finally, for distributed retrieval performance, Safaei [33,34], Le et al. [63], Galletta et al. [51] and Soltanmohammadi et al. [92] show that fragmentation-based strategies yield measurable query response time reductions compared to non-partitioned baselines, whereas Hadoop-based approaches [45,46] prioritize parallel processing throughput over indexing granularity. Some studies address the adaptation of interoperability strategies in scenarios where medical records are dispersed across multiple formats and locations [12].

To more clearly observe the strategies used to improve performance in each study, different tables (

Table 10. Software architectures and patterns.

Strategy	Study Reference	Benefit/Result
Use REST, modularization, and architecture distribution	[21]	A modular web/mobile platform enabling flexible management of cardiology records in a hospital pilot.
Microservices for component reuse	[40]	Data fusion architecture conceptually validated across 4 real-world cases (mental health, COVID-19, kidney disease, peritoneal dialysis).
Intelligent Software Agent Systems	[65]	Preliminary tests showed that multiple agents can simultaneously analyze the same set of DICOM images without interfering with each other.

,

Table 11. Data management and storage (cloud/distributed/sharding).

Strategy	Study Reference	Benefit/Result
Fragmentation	[33,34,63,92,93]	In [33], diamond index achieved 97.6% retrieval precision. In [34], hybrid fragmentation achieved 98.2% precision. In [63], the NSGA-G algorithm achieved the best execution time compared to other NSGA algorithms. Ref. [92] achieved  43–47% lower fetch times vs. non-partitioned tables.
Data Ingestion as a Service for Data Lakes	[14]	DIaaS achieved ingestion latencies of  148.1 µs/record (structured) and 234.2 µs/record (semi-structured), unifying heterogeneous ingestion.
Sharding and load balancing	[18]	Single-node MongoDB indexed each DICOM object in 18.9 ms versus 23.3 ms for Apache Lucene; the distributed scenario added 13.3 ms of network overhead.
Mix local and cloud storage transparently	[19]	The edge storage evaluation revealed QoS failure rates of up to 81–100% under heterogeneous workloads, identifying unresolved design deficiencies.
Using Cloud Aging and Machine Learning for Efficient Analytics	[24]	Proposed a conceptual 4-layer k-Healthcare model for e-Health/m-Health using smartphone sensors, without experimental validation.
Distributed memory usage	[38]	The loosely-coupled distributed EHR system achieves interoperability between healthcare centers and data retrieval via MPI backup, with no numerical metrics.
Cloud computing for data storage and processing	[42]	Prospective analysis of technology trends (publications, patents, press) on wearables in telehealth, with no original quantitative results.
Annotation storage as JSON instead of duplicating DICOM images	[49]	Functional collaborative platform for management and visualization of medical data/images with AI integration, with no quantitative metrics reported.
Using standards and cloud storage	[50]	Patient-centered PHR with a Raspberry Pi module and HL7 CDA/SNOMED CT standards for interoperability, with no quantitative performance evaluation.
Fragmentation of medical files	[51]	Multi-cloud system achieves scalable execution times as the number of providers increases from 6 to 12 (only  9000 ms difference), validating its feasibility.
Data integration with Apache NIFI	[68]	The platform demonstrated effectiveness in real-world cardiology, neurology, and radiology scenarios, improving data quality and interoperability.
Cloud storage for separation between application and data	[72]	Patient-centered record system with cloud-based metadata facilitates rapid retrieval of clinical data in emergencies (qualitative proposal).
Real-time integration of data from multiple sources using the cloud	[73]	Real-time data integration achieved a 20% reduction in average emergency response time in hospitals.
Document-oriented storage with binary embedding and metadata indexing	[93]	Unified storage of CXR image binaries with indexed metadata in a single MongoDB document.

,

Table 12. Distributed Computing and Big Data (Hadoop/MapReduce).

Strategy	Study Reference	Benefit/Result
Hadoop for parallel processing	[27]	The IoT and big-data remote monitoring framework achieved 99.96% accuracy in predicting patient physical activity
Distributed Medical Image Allocation Using Hadoop	[45]	The HDFS/Hadoop-based MIFAS system achieves high distributed storage reliability and better transmission performance than PACS for small files.
Fuzzy C-means and k-means hybrid clustering using Hadoop MapReduce	[46]	The Hybrid fuzzy k-means approach on Hadoop MapReduce achieves 96% accuracy in brain tumor detection and reduces execution time by 30%.
Using Big Data in Hadoop and Applying Parallelism in MapReduce	[80]	UHPr retrieves a patient’s complete medical profile error-free among more than 116.5 million medical fragments from 390,101 patients.

,

Table 13. Edge/IoT and Efficient Capture.

Strategy	Study Reference	Benefit/Result
Message size reduction in the IoT context	[1]	The SAREF4health ontology achieved a message size comparable to FHIR for ECG series, compared with standard SAREF (5 MB vs. 100 KB).
Using Kinect for localization with the ultrasound probe	[22]	The Kinect-based system achieved tracking accuracy of ±5 mm/±2°, comparable to the NDI Polaris locator (3.5 mm) at lower cost.
Local processing to reduce network and cloud load, event-driven instead of continuous streaming	[25]	Local-remote processing architecture in IoT/AAL that reduces transmission load and improves processing speed, with no quantitative figures.
Microcontrollers to optimize the use of blocks	[28]	Telemedicine prototype using Raspberry Pi that detects Tachycardia and Hyperkalemia via ECG R-peak analysis, with no quantitative metrics reported.
IoT to optimize data acquisition	[41]	Conceptual design proposal (MedDApp) for health monitoring using blockchain and IoT, without experimental validation or performance figures.
Hierarchical smartphone-like architecture improving the flow of information	[47]	Perspective article proposing migration of radiology IT architecture to a smartphone-like hierarchical model, with no quantitative data.

,

Table 14. Standards, Interoperability, and PACS.

Strategy	Study Reference	Benefit/Result
Load balancing and the use of WADO to improve response times	[23]	The web system achieved DICOM loading times of 0.90 s (50 CT slices) to 5.10 s (558 MRI slices) using WADO instead of C-MOVE.
Standardization with DICOM Structured Reports	[26]	Integrated 66,348 DICOM SR reports with PACS images; automatic reconciliation reduced inconsistent quality categories from 8 to 6 valid groups.
Interoperability between mHealth and eRecords applications	[30]	Qualitative study (surveys/interviews) identifying the need, opportunities, and the challenges of interoperability between mHealth apps and eRecord systems in Botswana, without quantitative metrics.
Workflow optimization through DICOM and PACS standardization	[37]	Qualitative system that improves data security and optimizes laboratory workflow through centralized authentication, with no quantitative figures
Use of heterogeneous standards and data integration	[85]	91.7% of healthcare professionals and 97.1% of IT professionals positively rated the proposed model as improving and simplifying clinical systems.
Hospital Systems Integration and Digitalization	[88]	Conceptual article that qualitatively discusses how hospital integration and digitalization improves efficiency, quality, and satisfaction.

,

Table 15. Blockchain, DLT, and distributed storage.

Strategy	Study Reference	Benefit/Result
Interoperability with Tangle	[12]	Achieved full interoperability and integration between PHR and EHR via a distributed IOTA-Tangle architecture with an HL7 FHIR API, without intermediaries.
Using blockchain for medical data management	[15]	BlockHR achieved data retrieval 20 times faster than client/server, although data write was 2.6 times slower.
BFT-PNT consensus protocols for distributing databases	[17]	The BFT-PNT protocol achieved a throughput gain of 1.4× (4 nodes) and 1.7× (8 nodes) versus traditional BFT schemes (Tendermint).
Decentralized architecture using blockchain	[55]	Proof of concept with 100 patients and over 1000 transactions demonstrated the feasibility of integrating heterogeneous PHRs into a decentralized, scalable blockchain.
Using Smart Contracts and Distributed Storage through Blockchain	[61]	The blockchain/IPFS system proved efficient and viable, with gas costs quantified for each smart contract operation.
Using IPFS to Solve Blockchain Block Capacity Limit	[64]	Blockchain-based PIE system achieved downloading 1 MB of medical history in an average of 10.1 ms, ensuring high security.

,

Table 16. AI/Deep Learning and Acceleration of Analysis.

Strategy	Study Reference	Benefit/Result
The use of an architecture for U-NET convolutional networks and deep learning	[36]	3D lung tumor segmentation with U-NET achieved 98.9% accuracy, 97.99% sensitivity, and a 97% Dice index.
Federated learning	[39]	Bibliometric review of 1548 articles (2010–2023) identifying trends and adoption gaps in Medical 4.0 technologies.
Deep learning algorithms to accelerate analytics	[56]	AI-based thermography system achieved 82.5% sensitivity and 80.5% specificity, outperforming mammography in women with dense breasts.
Using deep learning for hierarchical classification of images and tables	[59]	Multimodal deep learning model achieved a macro-average F1-score of 95.9% (87.5% specific) for identifying COVID-19.
Inclusion of cloud-based deep learning	[60]	Integration of deep AI and cloud computing improved diagnostic accuracy by 15–20% and reduced processing time by 60%.
Deep Learning and Massively Parallelization Using Multiple CPUs	[78]	Proposes a conceptual IoMIT and analytics architecture for early thyroid cancer diagnosis, with no quantitative validation results.
Using Fuzzy and Fractal Convolutional Neural Networks for Image Classification	[82]	The FDCNet model, evaluated on the BraTS dataset, achieves 98.68% accuracy in brain tumor detection and classification.
Cloud-driven data engineering and federated learning	[91]	The proposed architecture achieves 91% accuracy in semantic data mapping and reduces inter-institutional data retrieval time by 84%.
DenseNet121 for CXR pneumonia classification from MongoDB-stored images	[93]	Reported accuracy/precision/recall = 1.0 on MIMIC-IV CXR subset (230 normal/234 pneumonia, 80–20 split).

,

Table 17. Medical Image/Video Processing and Compression.

Strategy	Study Reference	Benefit/Result
Compression of DICOM videos with HEVC	[13]	HEVC compression achieved ratios of 10–15× for brain MRI and 25–27× for angiography, maintaining an SSIM of 0.8–0.9.
Artifact suppression algorithms, color maps, dynamic navigation	[31]	Combined artifact-suppression algorithms and color mapping in CBCT detected 100% of separated instruments, versus only 32.3% with periapical radiography.
Automatic adaptation of density scals in CT	[77]	Automatic grading correction improves CT/X-ray image contrast by up to 5 times and processes each image in under 1 s.
Apply Partial Loss Compression and Color Exclusion for Video Recovery	[81]	Automatic grading correction improves CT/X-ray image contrast by up to 5 times and processes each image in under 1 s.

and

Table 18. Architecture and Scalability.

Strategy	Study Reference	Benefit/Result
Microcontrollers to optimize block usage	[28]	Telemedicine prototype using Raspberry Pi that detects Tachycardia and Hyperkalemia via ECG R-peak analysis, with no quantitative metrics reported.
Microservices for component reuse	[40]	Data fusion architecture conceptually validated across 4 real-world cases (mental health, COVID-19, kidney disease, peritoneal dialysis), with no quantitative metrics.
Hybrid clustering (Fuzzy C-means, k-means) with Hadoop MapReduce	[46]	Hybrid fuzzy k-means approach on Hadoop MapReduce achieves 96% accuracy in brain tumor detection and reduces execution time by 30%.
Automatic adaptation of density scales in CT	[77]	Automatic grading correction improves CT/X-ray image contrast by up to 5 times and processes each image in under 1 s.

) are presented to classify and group the studies by similar strategies.

The strategies identified in Table 10, Table 11, Table 12, Table 13, Table 14, Table 15, Table 16, Table 17 and Table 18 reflect considerable technological diversity in optimizing the performance of medical databases. Data sharding, distributed computing using Hadoop/MapReduce, and the use of microservices and blockchain-based architectures are the most common approaches. Furthermore, the integration of artificial intelligence and deep learning techniques is emerging as a growing trend, complemented by edge computing strategies aimed at reducing the load on central repositories.

Regarding the relevance of traditional fragmentation in the context of cloud-native systems, both approaches are not mutually exclusive. While cloud platforms offer built-in sharding and replication mechanisms, explicit fragmentation strategies, such as those proposed by Safaei et al. [33,34], Galletta et al. [51], Le et al. [63], and Soltanmohammadi [92] remain valuable for optimizing DICOM metadata indexing, query performance, and data dissemination in domain-specific scenarios where generic cloud solutions do not account for the structural properties of medical imaging data.

From a technical standpoint, the works of Safaei & HabibiAsl [33] and Safaei [34] provide the most detailed evidence of fragmentation’s impact on DICOM data management. Vertical fragmentation reduced retrieval time by isolating frequently queried metadata attributes, while hybrid fragmentation further improved throughput by combining attribute-based partitioning with feature clustering. Complementing these efforts, Galletta et al. [51] apply a residue-based fragmentation scheme that splits MRI/DICOM files into redundant chunks distributed across multiple cloud storage providers, prioritizing data availability, reliability, and privacy through obfuscated metadata maps rather than query-time optimization. Meanwhile, Le et al. [63] combine vertical partitioning with a grid-based multi-objective genetic algorithm (NSGA-G) to derive row/column data configurations for hybrid DICOM storage, balancing storage space, monetary cost, and query response time. Although direct quantitative comparisons across studies are limited by the use of different data sets and environments, all four works report measurable reductions in query response time compared to non-fragmented baselines, reinforcing fragmentation as a technically justified strategy for large-scale medical image retrieval. Complementing these attribute-based fragmentation strategies, Soltanmohammadi et al. [92] propose a value-based hash partitioning scheme for structured patient records, combining MD5 hashing for anonymization with modulus-based segment assignment across native PostgreSQL partitions. Unlike Safaei’s vertical/hybrid fragmentation of DICOM metadata attributes or Le et al.’s grid-based multi-objective optimization of hybrid DICOM storage, this approach targets uniform load distribution of identifier-keyed transactional records rather than image-feature clustering. Taken together, these findings reinforce that fragmentation strategy selection should be guided jointly by data type and access pattern: attribute or feature-based fragmentation is better suited to imaging metadata retrieval, whereas hash-based partitioning is better suited to uniform load balancing of high-volume transactional patient records.

Synthesizing the strategies identified across Table 10, Table 11, Table 12, Table 13, Table 14, Table 15, Table 16, Table 17 and Table 18, three broad technological tiers emerge: (i) infrastructure-level strategies, including distributed computing, cloud storage, and fragmentation, which target scalability and query performance; (ii) integration-level strategies, such as microservices, pipeline architectures, and interoperability standards, which address data heterogeneity and system connectivity; and (iii) intelligence-level strategies, encompassing AI, deep learning, and federated learning, which focus on analytical capabilities and autonomous decision support. These tiers are not mutually exclusive; several reviewed works [24,37,38,40,41,46,56,60,80,91] combine approaches from two or more levels, reflecting the increasingly multi-layered nature of medical data management systems.

4.4. RQ4. What Data Sets Are Most Commonly Used to Evaluate Approaches Dedicated to Medical Data Management?

To evaluate proposed approaches in medical data management, researchers frequently rely on simulated data sets, synthetic workloads, real patient records, or prototype implementations in controlled environments. A common practice involves using simulated health records, which allow for the safe testing of data integrity and system behavior without compromising patient confidentiality (Ismail et al. [15]; Pedrosa et al. [17]). Similarly, synthetic data sets provide a means to assess scalability and flexibility under different data loads, as seen in the work of Sreepathy et al. [14] and Sondur et al. [19].

In some cases, real-world data sets are employed. For example, Pole and Shriram [13] used actual CT image sequences to test compression strategies, while Godinho et al. [26] conducted case-based evaluations in a cardiology lab. Other studies such as Ourahmoune et al. [22] built their own data sets from tests with healthy volunteers, reflecting a hybrid approach to data set creation.

A subset of studies relies on prototypes tested with sample data or pilot implementations in clinical settings to validate system feasibility (Saweros and Song [12]; Conte et al. [21]). However, some works notably lack formal benchmarking, relying instead on surveys or qualitative evaluations (Dhayne et al. [20]; Spinsante et al. [25]).

Table 19. Benchmark or data set.

Benchmark or Data Set	Study Reference	Classification
TCIA	[14,33,36,46,78]	Unstructured (DICOM images with structured metadata)
Indian Stock Market Dataset	[14]	Structured
NSE Stocks Data	[14]	Structured
Climatic Information by Openweather	[14]	Semi-structured (API/JSON data)
Tuberculosis dataset from https://tbportals.niaid.nih.gov/ (accessed on 11 May 2026)	[18]	Structured
Knee Ultrasound Learning Database	[22]	Unstructured (ultrasound images/video)
Vietnam Medic Medical Diagnostic Center Dataset	[23]	Unstructured (DICOM images)
Dataset from a hospital affiliated with the University of Aveiro in Portugal	[26]	Unstructured (DICOM images and ECG signals)
Physical Activity Monitoring benchmarked database	[27]	Semi-structured (sensor/wearable time series)
CARDIODAT of PTB	[28]	Unstructured (ECG signals)
PhysioNet	[28]	Unstructured (physiological signals)
CBCT images of mandibular molars, source not mentioned	[31]	Unstructured (CBCT images)
RadiologyNET dataset from the Clinical Hospital Center Rijeka	[32]	Semi-structured (images + DICOM tags + narrative diagnoses)
Images of Tabriz Behbood Hospital	[33]	Unstructured (DICOM images)
PACS of the Salah Azaiez Institute (Tunisia)	[36]	Unstructured (PACS DICOM images)
GNU Health	[38]	Structured (clinical records in a relational database)
Multicenter PSG registries of the German Sleep Society	[44]	Unstructured (polysomnography signals)
DICOM data from the IRCCS case study ‘Bonino Pulejo’	[51]	Unstructured (DICOM files)
Bucharest clinic database	[55]	Structured (tabular clinical records)
King Khalid University Hospital Dataset	[59]	Unstructured (radiographic images)
Rashid Hospital Dataset, Dubai	[59]	Unstructured (radiographic images)
CTColonography	[63]	Unstructured (DICOM images)
Dclunie	[63]	Unstructured (DICOM images/files)
Idoimaging	[63]	Unstructured (DICOM image repository)
LungCancer	[63]	Unstructured (DICOM images)
MIDAS	[63]	Unstructured (DICOM images)
CIAD	[63]	Unstructured (DICOM images)
Simulations in Hyperledger Fabric	[64]	Semi-structured (blockchain transactions/blocks)
BigPicture	[67]	Semi-structured (histopathology images + clinical metadata)
MAGIX	[69]	Semi-structured (CT/DICOM study with RDF annotations)
Digital Knee X-ray images (generated)	[75]	Unstructured (radiographic images)
Samsun Dental Hospital Radiographic Database	[76]	Unstructured (intraoral radiographic images)
Images of pig organs with metal fragments	[77]	Unstructured (experimental CT images)
Thyroid Digital Images Database	[78]	Unstructured (medical images)
Garavan Institute Database	[78]	Unstructured (medical images)
MIMIC-IV CXR (subset)	[93]	Semi-structured
Simulated/synthetic patient records	[92]	Structured

is presented, relating the different benchmarks or data sets to the works that use them.

As can be seen in Table 19, the most used data set to evaluate medical data management systems is TCIA, although it is noted that the vast majority of the works use different data sets. It should be noted that some data sets listed in Table 19, such as stock market or climate datasets, appear unrelated to the medical domain. These were included in the reviewed studies as part of multi-domain ingestion pipeline evaluations, where medical data was processed alongside heterogeneous non-medical sources to demonstrate system scalability and flexibility. Their presence in the table reflects the scope of those specific works rather than a thematic misalignment with this review.

To improve reproducibility, future studies should adopt a minimum set of reporting standards, including: the explicit identification of the data set used (size, modality, and source), the hardware and software environment, and the performance metrics evaluated (e.g., query response time, throughput, storage overhead). Adopting widely recognized data sets such as TCIA as a common baseline would further facilitate meaningful comparisons across approaches.

The inclusion of the MIMIC-IV CXR subset (Dubey and Saxena [93]) further substantiates the prevalence of the MIMIC family of datasets noted in the concluding remarks of this review, and partially addresses the current scarcity of MIMIC-based entries in Table 18.

Beyond reporting standards, the establishment of problem-specific benchmark datasets would significantly advance the field. For instance, a unified dataset for DICOM metadata retrieval performance, another for sensor-to-repository transmission latency, and another for heterogeneous data integration could serve as common baselines across studies. TCIA represents a step in this direction for imaging data, but its scope does not cover the full range of problems addressed in this review, underscoring the need for community-driven efforts to define and maintain standardized evaluation corpora for each research question.

4.5. RQ5. How Is Medical Information Transmitted from Sensors Included in Wearables and Imaging Machines to a Central Data Repository?

The transmission of medical data from wearable sensors and imaging machines to central repositories varies depending on the type of device and system architecture employed. Wearable sensors, which collect physiological data in real time, often transmit information through mobile applications or wireless communication protocols to intermediary devices or directly to cloud-based systems. For example, Ullah et al. [24] explicitly mention the use of wearable sensors for data capture, while Latif et al. [16] describe a mobile-based system that handles basic health data inputs via SMS or mobile apps, facilitating lightweight communication in resource-limited settings.

In more advanced scenarios, ambient and spatial sensors like those used by Spinsante et al. [25] and Ourahmoune et al. [22] capture motion, temperature, or ultrasound data, which are then transmitted via local networks to centralized platforms for processing. These systems are typically part of IoT-enabled architectures that emphasize interoperability and continuous monitoring.

For diagnostic imaging systems such as MRI, CT, or X-Ray machines, transmission often relies on established standards like DICOM, which supports structured file ingestion from PACS. This is evident in the work of Almeida et al. [18], who describe the ingestion of DICOM files from PACS servers, and in Dhayne et al. [20], who focus on standard communication protocols for MRI and X-Ray modalities. Additionally, Godinho et al. [26] illustrate integration of ECG signals and DICOM images in a cardiology lab environment, showing the convergence of multi-modal data transmission.

Some systems opt for broader ingestion pipelines, supporting diverse data sources including sensors, administrative systems, and electronic health records (Sreepathy et al. [14]; Pedrosa et al. [17]), reinforcing the trend toward heterogeneous data integration.

Based on the reviewed studies, three broad categories of sensors were identified: (i) wearable and IoT sensors, which capture physiological variables such as ECG, SpO₂, and body temperature and typically transmit data via mobile or wireless protocols [1,12,15,19,24,27,38,40,41,44,50,55,58,60,62,68,72,73,80,88,91]; (ii) environmental and ambient sensors, such as motion or temperature detectors used in assisted living contexts [22,25,53]; and (iii) imaging device sensors, including CT detectors, MRI coils, and ultrasound transducers, which generate high-volume DICOM data transmitted through PACS infrastructures [13,17,18,21,23,26,29,31,32,33,34,36,45,46,47,49,51,59,60,61,63,64,65,69,71,72,75,76,77,78,79,82,86]. This classification highlights that each sensor category interacts differently with DICOM standards and medical databases, and that integration strategies must be tailored accordingly.

Table 20. Summary of device types, communication media, protocols, and standards used in the selected studies.

Study Reference	Device/Sensor Type	Medium	Protocol	Standard/Data Model
Syed et al. [27]	IMU sensors for athletes, heart rate, temperature, heartbeat monitors, blood pressure sensor	Wi-Fi	Not specified	Not specified
Le et al. [63]	Medical imaging modalities (CT, MRI, Ultrasound, etc.)	Cloud computing environment	Not specified	DICOM
Rinty et al. [38]	Remote-Local healthcare node	Wi-Fi	MPI, SSH, HL7 communication	HL7 CDA
González Bermúdez et al. [40]	Smartphone (embedded sensors, user input)	Wi-Fi	HTTP (RESTful communication)	HL7 FHIR
González Bermúdez et al. [40]	IoT health devices	Bluetooth/NFC/Wi-Fi	MQTT/REST API	HL7 FHIR
Wang & Nurcahyo [41]	Wearable sensors (IoT devices)	Mobile network/ Cloud/Blockchain	Smart contracts over Ethereum	FHIR, ICD-10, EDIFACTS, DICOM
Wang & Nurcahyo [41]	Blockchain nodes (BigchainDB + IPFS)	Internet/P2P decentralized network	Ethereum protocol	NoSQL blockchain
Lee et al. [64]	Medical imaging devices (MRI, CT, X-ray, endoscopy)	IPFS peer-to-peer network	IPFS protocol for distributed file transfer	DICOM
Vázquez-Ingelmo et al. [49]	Echocardiographic, MRI, CT imaging devices	Internet (web-based platform, client–server)	HTTP (web API calls), Django ORM over TCP/IP	DICOM
Song et al. [50]	Raspberry Pi 3 + e-Health Sensor Shield V2.0 (biosensors: pulse rate, SpO2, respiration, temperature, glucose, ECG, GSR, blood pressure, accelerometer, EMG)	Local connection between sensor board and microcontroller	Not specified	HL7 CDA
Galletta et al. [51]	MRI scanner	Internet (multi-cloud storage network)	HTTP over TCP/IP	DICOM
Beier et al. [44]	Polysomnography	Internet (HTTP-based connection over TLS encryption)	REST API (HTTP/HTTPS)	EDF/EDF+
Praveenkumar et al. [29]	CT and MRI scanners	Wireless spectrum (Cognitive Radio)	QAM	DICOM
Saweros & Song [12]	IoT sensors (SPO2, ECG, BP, body temperature)	Internet/Wireless (via smartphone connection)	Not specified	HL7 FHIR, C-CDA
Czelusniak et al. [65]	Tomography and MRI scanners	Local computer network	FIPA-compliant agent communication	DICOM
Cernian et al. [55]	Wearable sensors (Fitbit fitness tracker)	Internet/Cloud (Fitbit web API)	HTTP/REST	HL7
Pole & Shriram [13]	MRI scanner	Not specified	Not specified	DICOM
Estrela et al. [31]	Cone-beam computed tomography (CBCT) scanner	Local workstation (desktop computer)	Not specified	DICOM
Schreiweis et al. [68]	EEG devices (“Dreem 2”, “V-Amp”)	S3 object storage (via internal network)	Not specified	BIDS
Vega et al. [69]	CT, MR scanners	Web (Internet)	WebSocket	DICOM, RDF/XML
Ismail et al. [15]	Health sensors and medical devices	Internet/local network	PBFT	Blockchain ledger
Pedrosa et al. [71]	MRI scanner	Distributed ledger network/Distributed file systems (DLT + IPFS)	DICOM transfer	DICOM
Yang et al. [45]	MRI, CT, Ultrasound, PET, Endoscopy, Mammogram, DR, CR	Ethernet/TANET (100 Mbps academic network)	HTTP, HDFS block protocol	DICOM
Alyami et al. [72]	Wearable body sensors and Imaging machines	Internet/Cloud	HTTP	HL7 CDA y DICOM
Munagandla et al. [73]	Wearable health devices	Bluetooth/Wi-Fi	MQTT/HTTPS (implied for real-time IoT data)	HL7
Gornale et al. [75]	X-ray machine (PROTEC PRS 500E)	Not specified	Not specified	DICOM
Pedrosa et al. [17]	X-Ray, MRI, CT imaging devices	Local network	BFT-PNT	DICOM
Chahal & Pandey [46]	MRI scanner	HDFS over local network	MapReduce	DICOM
Napravnik et al. [32]	Medical imaging modalities (CT, MRI, Ultrasound, etc.)	PACS	TCP/IP	DICOM
Safaei & HabibiAsl [33]	Medical imaging devices (MRI, CT, X-ray, PET/CT, Mammography)	Local hospital network	Not specified	DICOM
Safaei [34]	MRI, CT, X-ray, PET, fMRI, Radiography	Not specified	Not specified	DICOM
Almeida et al. [18]	CT scanner	Local network	TCP/IP	DICOM
Deniz & Kaya [76]	Intraoral phosphor storage plate system	PACS	Not specified	DICOM
Sondur et al. [19]	IoT/edge devices	Wireless	Not specified	DICOM
Khoroshun et al. [77]	CT scanner	Local computer	Not specified	DICOM
Huda et al. [58]	Wearable and patient monitoring devices	Wi-Fi	TCP/IP, WPA2	HL7
Borovska et al. [78]	Ultrasound/MRI/CT scanner	Internet	HTTP (implied via cloud services)	DICOM
Enzmann et al. [47]	Imaging modalities (CT, MRI, Ultrasound)	PACS	Not specified	DICOM
Zimmerer & Gellrich [79]	3D C-arm	Local wired connection	Not specified	DICOM
Satti et al. [80]	Medical IoT sensors	Wi-Fi	TCP/IP	HL7, FHIR, DICOM
Podchashynskyi et al. [81]	Video camera	Not specified	Not specified	DICOM, JPEG
Elloumi et al. [36]	CT scanner (Computed Tomography)	PACS	DICOM transfer	DICOM
Gleiss & Lewandowski [53]	Bed sensor system	Local hospital network	REST API (HTTP/HTTPS)	HL7, DICOM, IHE
Molaei et al. [82]	MRI scanner	Not specified	Not specified	DICOM
Moreira et al. [1]	ECG wearable (Shimmer3 ECG unit)	Bluetooth, Wi-Fi	JSON-LD over HTTP	SAREF4health ontology
Rönnau et al. [85]	CT, physiological sensors (body temperature, blood pressure, heart rate)	Not specified	DICOM file access via PixelMed Java DICOM Toolkit	DICOM
Leif & Leif [5]	Flow cytometer	Not specified	XML Schema Definition	CytometryML
Conte et al. [21]	ECG, ultrasound imaging	Local network	RESTful API	DICOM
Bracciale et al. [86]	Medical imaging modalities (MRI, CT, X-ray, ultrasound)	Local network to PACS	TCP/IP	DICOM
Tiriteu et al. [88]	IoMT devices and sensors	Wi-Fi/Bluetooth	HTTP	DICOM
Althenayan et al. [59]	CXR (Chest X-ray) imaging system (Optima XR240amx)	PACS	DICOM transfer	DICOM
Shakor & Khaleel [60]	Wearable sensors, MRI, CT, X-ray scanners	Cloud infrastructure	TCP/IP	DICOM
Mohsan et al. [61]	MRI scanner	Internet	IPFS protocol	Blockchain ledger
Pendyala [91]	IoT-enabled medical devices	Cloud (AWS)	Event-driven (Kafka)	HL7 FHIR, DICOM
Khan et al. [62]	Medical IoT devices	Wireless network	Not specified	DICOM
Ourahmoune et al. [22]	Kinect sensor, FlexiForce force sensors	USB	TCP/IP	Not specified
Nguyen et al. [23]	CT, MRI, Ultrasound, X-ray	Internet	DICOM protocols	DICOM
Ullah et al. [24]	Smartphone	3G/4G/Wi-Fi	TCP/IP	IEEE 802.11/b/g/n
Spinsante et al. [25]	Home automation sensors	CAN bus	HTTP	JSON
Godinho et al. [26]	Echocardiography	Local network	DICOM Storage Service	DICOM SR, HL7

presents an in-depth analysis of all the works that include or mention how the various devices or sensors communicate and collect data. Some works include more than one form of transmission, or the topic addressed encompasses various sources and forms of communication; in these cases, those described in more detail in each report are included.

In summary, the transmission of medical information from sensors to central repositories involves a combination of mobile and wireless technologies for wearables, and standardized DICOM-based communication for imaging devices. These are often integrated into broader IoT and CPS infrastructures that facilitate real-time data exchange and centralized analysis.

Table 20 shows that DICOM is the predominant standard for transmitting medical image data, adopted across virtually all acquisition modalities analyzed. Wi-Fi and the Internet are the most widely used communication channels, while protocols such as HTTP/REST and HL7 FHIR are prevalent in clinical interoperability environments. Also noteworthy is the growing presence of decentralized technologies such as IPFS and blockchain for secure data transmission, as well as the use of Bluetooth and mobile networks in wearable devices for remote patient monitoring.

4.6. RQ6. What Strategies Are Used for Storing and Viewing Images in DICOM Format in Heterogeneous Repositories, Such as Electronic Medical Records?

In heterogeneous medical repositories, such as EHRs, the integration of DICOM images presents challenges related to format diversity, access control, and data linkage. Various strategies have been proposed to address these challenges and ensure seamless storage and viewing of imaging data. A common approach is the conversion and compression of DICOM images into lighter formats to facilitate integration and visualization. Pole and Shriram [13], for instance, describe a method where CT DICOM scans are converted to YUV format for optimized compression and easier handling within EHRs. Another prominent strategy involves web-based DICOM viewers, allowing clinicians to access images directly from heterogeneous platforms. Nguyen et al. [23] demonstrate the feasibility of viewing DICOM files on smartphones through browser-accessible tools, improving usability and mobility.

Some systems integrate DICOM storage into distributed database structures, such as Cassandra, enabling scalable and secure management. Almeida et al. [18] combine this with access management mechanisms to support diverse medical data formats across a unified repository. Efforts have also focused on linking structured and unstructured medical data, including DICOM, using metadata and semantic processing techniques. Dubey and Saxena [93] illustrate a related pattern at the NoSQL level, embedding CXR image binaries alongside descriptive metadata within a single MongoDB document, simplifying retrieval without requiring joins across separate structured and unstructured stores, though notably storing the images as JPEG rather than preserving the native DICOM container. Conte et al. [21] employ semantic rules to connect structured records with imaging information, while Godinho et al. [26] demonstrate a warehouse model that integrates DICOM with other clinical data streams. In blockchain-based systems, as described by Ismail et al. [15], DICOM images are stored via IPFS, and patient records are securely managed in a decentralized architecture, promoting traceability and access control. Finally, in low-resource contexts, simpler techniques such as text-based encoding of medical information, as seen in the work of Latif et al. [16], are used to maintain some level of interoperability between structured data and image-derived summaries.

Table 9 describes a compendium of viewers and PACS systems found in the studies included in this research.

In summary, strategies for handling DICOM in heterogeneous repositories include format conversion, web-based viewing, distributed and blockchain-backed storage, metadata linkage, and integrated data warehousing. These approaches ensure flexible, scalable, and accessible image management across complex medical data ecosystems.

Table 9 shows that Dicoogle is the most frequently referenced PACS system in the analyzed studies, standing out for offering both storage and visualization capabilities. However, a significant proportion of the studies opt for developing their own implementations, reflecting the need for solutions adapted to specific environments and requirements. Tools such as Orthanc, Cornerstone.js, and XNAT also appear as established alternatives, while several specialized viewers, such as RadiAnt or MicroDicom, are used exclusively for visualization. This diversity underscores the lack of a unified standard in the adoption of PACS systems, representing an area of opportunity for interoperability in heterogeneous clinical environments.

4.7. RQ7. What Are the Main Challenges and Areas of Opportunity in the Current Medical Database Landscape for Future Developments?

Future developments in medical databases are shaped by several recurring challenges and promising research directions. A major area of opportunity lies in enhancing interoperability and infrastructure scalability, which are critical for integrating diverse data sources such as wearables, imaging devices, and hospital systems. Studies highlight the importance of IoT integration and smart contract support to automate and secure data handling (Saweros and Song [12]; Latif et al. [16]).

Another recurring challenge is real-time data ingestion and domain-specific adaptability, particularly in systems dealing with high-volume sensor inputs or heterogeneous medical information. Sreepathy et al. [14] advocate for real-time pipelines and modular data connectors tailored to specific clinical domains. Similarly, Ismail et al. [15] emphasize the need to support a broader spectrum of clinical and administrative data in future solutions.

In terms of performance, there is a recognized need for optimization in indexing, fault tolerance, and hybrid cloud deployments, as highlighted by Almeida et al. [18] and Pedrosa et al. [17]. These improvements aim to balance efficiency with reliability in large-scale or distributed environments.

Several researchers also point to the lack of unified standards and data quality mechanisms, which hinder cross-platform integration and intelligent analytics (Dhayne et al. [20]; Spinsante et al. [25]). Enhancing automation, user customization, and AI support, especially for diagnosis or anomaly detection, has been proposed to address these gaps (Nguyen et al. [23]; Godinho et al. [26]).

Finally, expanding national-level platforms, remote care systems, and patient-centered designs is seen as a frontier for widespread adoption, particularly in contexts aiming for inclusive and scalable health monitoring solutions (Conte et al. [21]; Ullah et al. [24]).

In summary, the future of medical databases is defined by the need for more interoperable, real-time, and intelligent systems. This includes integrating IoT technologies, optimizing architectures for scale and reliability, and aligning infrastructure with patient-centric care and clinical decision support.

The rise of federated learning suggests a gradual shift from purely centralized repositories toward hybrid edge-cloud architectures, where data remains near its source for training purposes while only aggregated model updates are shared centrally. This evolution does not eliminate the need for central repositories, but redefines their role as coordination and governance hubs rather than sole storage points.

The tension between heterogeneous data integration and patient privacy mandates such as GDPR and HIPAA remains a critical challenge. Emerging approaches such as differential privacy, trusted execution environments, and pseudonymisation protocols offer promising paths to reconcile research utility with de-identification requirements, though their adoption in large-scale, multi-institutional deployments is still limited and represents a priority area for future development.

5. Conclusions

Medical database platforms are steadily converging on hybrid, distributed architectures that blend mature relational engines with NoSQL stores, blockchain ledgers, and emerging decentralized-EHR frameworks. This combination provides the elasticity and fault tolerance needed to accommodate highly heterogeneous data (structured clinical notes, real-time physiological waveforms, genomic sequences, and ever-growing DICOM image libraries) while preserving end-to-end auditability. Performance gains are now achieved less through monolithic hardware upgrades and more through in-memory processing, container-oriented microservices, and selective migration to public or hybrid clouds. These strategies compress query latency to near–real time, enabling bedside analytics, continuous ICU monitoring, and AI-assisted decision support. Although DICOM remains the de facto standard for imaging, it is increasingly wrapped in DICOMweb services, RESTful PACS gateways, and object-storage back ends (e.g., S3-compatible buckets), with dynamic transcoding pipelines and HL7-FHIR bridges ensuring multichannel visualization and cross-institution data exchange. On the acquisition side, data from wearables and imaging modalities traverse Bluetooth LE, Wi-Fi 6, 5G, or Zigbee links, funnelled through edge gateways that apply preprocessing, encryption, and rule-based filtering before forwarding to the central repository, offloading core resources, and safeguarding privacy during intermittent connectivity.

Despite these advances, several obstacles persist. Semantic interoperability is still hampered by fragmented ontologies and inconsistent metadata, underscoring the need for unified vocabularies and automated FAIR-compliance workflows. The accelerating adoption of deep-learning pipelines demands infrastructures that can scale horizontally without compromising transactional integrity or regulatory alignment. Privacy and ethics remain front-of-mind, propelling research into federated learning, trusted-execution enclaves, and differential-privacy controls that satisfy both GDPR and HIPAA mandates. Benchmarking, meanwhile, is dominated by the MIMIC and Open-I data sets; broader, openly accessible corpora (particularly in oncology, cardiology, and telemedicine) are essential for rigorous, generalizable evaluations. Looking forward, edge-to-cloud topologies, energy-aware scheduling, and standardized performance metrics (latency, throughput, carbon footprint) will frame the next wave of innovation. Successfully addressing these challenges will pave the way for truly interconnected, patient-centric health ecosystems in which clinical information flows as seamlessly as it is generated, powering safer, faster, and more equitable care.